Perseverance Pays Off: Examining Behavioral Engagement in Learning Management Systems and Predicting Performance with Polygonal Curve Analysis

High and regular behavioral engagement is important for high academic performance in college. However, students vary greatly in their ability to regulate their learning and many students struggle with efficiently managing the time and effort of learning activities. Digital trace data from learning management systems (LMS) can be used to examine students’ learning behaviors in authentic classes and to identify students at risk of performing poorly. However, challenges remain regarding the behavioral measures and timeframes for this purpose. We analyze digital trace data and use broad measures of behavioral engagement in four introductory chemistry courses at one US university to (a) identify high and low achieving students based on their trajectories of learning activities on the course LMS, (b) examine common patterns of students’ learning activities regarding the amount and consistency across the 10-week course, and (c) examine the earliest possible time frame to identify struggling students with a satisfactory prediction accuracy based on their learning behavior. Polygonal curve analysis with elastic dissimilarity measures could identify high or low achieving students based on their digital trace data with an accuracy above 75%. K-means cluster analysis revealed four clusters of study patterns. 13-27% of students in each class had high and consistent amount of click activities, whereas 13-21% had low amounts and low consistency in click activities. Cluster membership was closely related to course performance. Students who were likely to struggle in the course could be identified with polygonal curves and ordinal regression based on click activities in weeks 1-4.